An algorithm to detect non-background signals in greenhouse gas time series from European tall tower and mountain stations

Abstract. We present a statistical framework for near real-time signal processing to identify regional signals in CO2 time series recorded at stations which are normally uninfluenced by local processes. A curve-fitting function is first applied to the detrended time series to derive a harmonic describing the annual CO2 cycle. We then combine a polynomial fit to the data with a short-term residual filter to estimate the smoothed cycle and define a seasonally-adjusted noise component, equal to two standard deviations of the smoothed cycle about the annual cycle. Spikes in the smoothed daily data which rise above this 2σ threshold are classified as anomalies. Examining patterns of anomalous behavior across multiple sites allows us to quantify the impacts of synoptic-scale weather events and better understand the regional carbon cycling implications of extreme seasonal occurrences such as droughts.